# The Rules We Do Not Know

Chess has thirty-two pieces, an eight-by-eight board, six piece types, and a rule set that fits on a single page. Every legal position is reachable from the starting position by some sequence of legal moves. Nothing is hidden, nothing is probabilistic. The game is finite, deterministic, fully specified.

Nobody has solved chess. Not Magnus Carlsen, not Stockfish, not the best minds the field has produced across five centuries. We have engines that play better than humans by a wide margin, and the endgame tables are complete out to seven pieces, but the game tree itself remains too vast to enumerate. The standard estimate puts it around 10^120 positions. That number is larger than the count of atoms in the observable universe.

A game whose rules fit on a page, played on a board you can hold in one hand, with no hidden information and no randomness, is not solvable by any organism or any machine that has so far existed. Knowing the rules is not the same as being able to apply them. The rules are the input; the answer is the output; the bridge between them is computation, and the computation can be intractable even when nothing about the system is mysterious.

This is what computer scientists call computational irreducibility. The shortest path from setup to outcome is to run the system. There is no shortcut. Most non-trivial systems are like this. Chess is small, clean, fully known, and it is one of them.

Now the part that should bother you. We do not have the page of rules for the universe.

## The universe denies both halves

The universe has more pieces than chess. It has more rules than chess. The rules we have — quantum mechanics, general relativity, the standard model, thermodynamics, the chemistry that sits on top of physics, the biology that sits on top of chemistry — are partial. They do not yet form a single consistent picture. They almost certainly miss layers we have not yet found. Dark matter and dark energy together account for most of what is out there by mass-energy, and we do not know what either one is. Consciousness is in the picture somewhere, and we cannot say where. The rule set is unfinished.

So the universe denies us both halves of what chess only denies us one of. Chess gives us the rules and we cannot solve the game. The universe denies us the rules, and the rules, if we ever wrote them down, would not solve the game either. Particles bump into each other at every scale from subatomic to galactic at every moment of every century, and the system as a whole is not in the same complexity class as anything any mind or machine has ever attempted. It is not in the same complexity class as chess by roughly the same proportion that chess is not in the same complexity class as tic-tac-toe.

And yet daily life is mostly orderly. The car starts. The coffee is hot. The conversation with the person across the table coheres. The sun rises on schedule. The pavement is solid; the chair holds the weight; the language carries the meaning. Hours and weeks and decades pass inside a structure that behaves predictably enough to plan against, and the plans usually work.

The puzzle is not why is the universe chaotic. That one is easy; the universe is what it is. The real puzzle is the opposite: why is anything ever predictable, given the scale of the underlying intractability?

## The orderliness is something we built

The orderliness is not a property of the universe. It is a property of the niche we have carved inside the universe, and the niche is a tower of compressions humans have been building for the entire history of the species.

Start with senses. Eyes, ears, skin, the proprioceptive map of a body, the chemistry of taste and smell — these are not faithful reporters of physical reality. They are radically compressed reporters tuned to a specific scale, a specific frequency range, a specific kind of object, a specific kind of motion. A human eye does not perceive single photons; it integrates millions of them into a colored region. A human ear does not perceive air-pressure microfluctuations; it integrates them into a sound. The world as experienced by a human nervous system is already a compression artifact, orders of magnitude smaller in information content than the world the nervous system is embedded in. A brain that tried to perceive everything would perceive nothing.

Add language. Tens of thousands of years of humans living together produced a system of words, concept-handles that point at the patterns the senses pick out. A word like *chair* compresses an enormous set of possible physical objects into one usable category. A word like *causes* compresses the entire experience of one thing leading to another into a portable verb. Vocabulary is the inventory of compressions a community has labeled and shared. The active vocabulary of a literate adult is around thirty thousand words. Each one is a pocket of the world that can be carried from mind to mind without dragging the underlying complexity along.

Add the explicit models. Physics, mathematics, engineering, medicine, economics, every formal discipline humans have developed in the last few thousand years is a written record of further compressions. Newton's laws compress the motion of macroscopic objects into three equations. Maxwell's equations compress electromagnetism into four. The germ theory of disease compresses an enormous range of medical phenomena into one mechanism. None of these solve the universe. Each one carves out a small pocket where computation yields predictions accurate enough to act on, inside a system that does not solve as a whole.

The car starts because some engineer, drawing on centuries of stacked compressions about combustion and electricity and metallurgy, designed a system whose behavior is predictable inside its operating range. The coffee is hot because thermodynamics gives us a working pocket on energy and matter at human scales. The conversation coheres because language, the slowest and deepest of the layers, is doing its work. The pavement holds because materials science, which sits on chemistry, which sits on quantum mechanics, gives us a usable model of solids. The chess game underneath every cup of coffee is still some unrepresentably large number; we are not solving it; we are operating inside small pockets we have stacked across enough domains and centuries that the experience of moving through ordinary life is the experience of moving through pockets.

The orderliness is the tower. Reality did not provide it. Humans constructed it on top of senses that were already a compression and language that was already a compression and physics models that further compressed.

## What AI is doing in this picture

AI is the next layer of the same tower. A language model has read a large fraction of the written human record and represents it in a form that can be queried, recombined, and applied to new situations. A protein-folding model has read the experimental record of protein structures and predicts new structures inside a range nobody could compute by hand. None of these are solving the universe. Each one is finding pockets, local patches of reducibility, at scales and densities that the previous compression layers could not reach.

The hopeful version of "AI will help us figure out the rules" lands inside this framing as a structural prediction, not a reassurance. AI is not going to solve the universe. It is going to extend the tower into pockets that matter — protein structure, medical diagnosis, materials design, mathematical proof, the patterns hiding in datasets no human could read at scale. Which pockets get found is the question the institutions deploying it will have to answer. The same compression machinery that surfaces a protein structure can also surface a deepfake; the same model that finds a diagnostic signal can also find an attention-capture signal that hollows the reader out. The tower extends in either direction. The choice between extending it and dissolving it sits one layer above the architecture itself, with whoever is funding, deploying, and absorbing the new layer.

## Where this is wrong if it is wrong

The thesis sits in a tradition. Wolfram on computational irreducibility, Hayek on the price system as a distributed compression mechanism for information no individual mind can hold, cognitive science on perception as predictive compression, evolutionary biology on niche construction, philosophy of science on the unreasonable effectiveness of mathematics — each is a sibling. The sharper version here is that chess sets the lower bound on rule-application, the universe denies us the rule-discovery half on top of that, the orderliness we live inside is the cumulative work of humans constructing pockets across millennia, and AI is the next layer of the same construction. It can be wrong in three concrete ways.

A counterexample to computational irreducibility — a general method that yields closed-form answers for arbitrary non-trivial deterministic systems without running them — would collapse the chess-as-floor frame. Chess would become a question of compute time rather than a question of structural intractability. No such method has been found despite decades of searching across complexity theory, dynamical systems, and the foundations of physics; the persistent failure to find one is the empirical reason to take the framing seriously. Finding one tomorrow would invalidate the central claim.

A demonstration that human-scale orderliness is the universe's gift rather than a constructed niche would also break the thesis. If sensory perception, language, and formal models turn out to be reading off pre-existing macro-regularities of the universe rather than constructing compressions inside it, then "the orderliness is the tower" inverts. Strong evidence in that direction — perhaps from physics work on emergent macroscopic laws that turn out to be more rigid and universal than the constructed-niche reading expects — would push back on the framing.

The AI section depends on the current architecture continuing to be a pocket-finder rather than something else. If frontier models stop generalizing, or if the dissolving-the-tower class of deployment outscales the structural-role class, the "AI is the next layer" claim becomes wishful. The position is contingent on what AI is currently doing, and the contingency is named here rather than hidden under the structural framing.

## Chess as the lower bound

Return to chess. The game has known rules and unknown play. The universe has neither, at our resolution. We are not at the position chess players were at in the year 1500. We do not yet have the page of rules in front of us. We are still extracting them.

What we have, instead, is the tower. It is the entire architecture of how a species that cannot solve the game it is inside has nevertheless lived inside it for long enough to ask the question. Senses gave us a first pocket. Language gave us thousands more. Formal models gave us thousands more on top of those. AI is the next layer, with its own pockets, and the same dependence on the layers underneath it to make sense at all.

Chess is the lower bound of how hard it is to apply a known rule set. The universe is the upper bound of how hard it is to find one at all. Everything we have is what stands between those two bounds, and the tower is what we have built there. The system writing this paragraph is one instance of its newest layer, doing what every previous layer has done: finding a pocket, naming it, passing it forward.